decision model · v1 set (earlier version)

GPT-6 LunaDecisions API

Earlier version

Results f24d1f41, updated 16 November 2026, 22:20 UTC; scored by GNU Backgammon 1.08, 2-ply.

How we called it

Model ID
gpt-6-luna
Provider or endpoint
OpenAI Decisions API
Date run
Missing
Price
$0.1 / $0.0 per million tokens in / out; API price; price date 2026-10-08
Model type
decision model
Weights
Missing
Runs on
API
Reasoning setting
none (decision model)
Confidence source
decision-model probability
Rules text
short (the v1 rules text; the same stored request text for every model)
Option groups
not recorded; no decision in this data was split
Sampling settings
not sent (decision request)
Notes
Missing

Missing B7 fields: weights, option_groups, date_run

Results in v1 set (earlier version)

Earlier version: plays were listed in the runner's order, which tends to put the best play early. When Gemini 4 Argon answered the same 1,000 positions with the plays shuffled, its error rate changed by +1.9 (95% interval −0.2 to +4.2), so the effect is not proven.

Error rate (PR)
35.031.0 to 39.4
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
44.2%
Blunder rate
25.3%
P95 equity loss
0.316
Mean loss, worst 5%
0.541
Median / P90 s per decision
0.22 / 0.27
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
54.3 (95% interval 51.7 to 57.0); risks found 20.0% (71 / 355), right safety 88.6% (327 / 369)
Risk judgement, margin 0.02
54.1 (95% interval 51.5 to 56.8); risks found 19.5% (62 / 318), right safety 88.7% (360 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

54.3risk judgement, margin 0.001

right risks71
wrong risks33
needless risks42
missed risks251
right safety327

Hit risk only: being hit. Gammon risk and priming are not in it.

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact40.5641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60